The first thing you notice, sitting across from him in a small café two streets back from the river, is how little of his attention is performing for the camera he knows is there. He is twenty-four. He runs an AI agency from Chiang Mai. He has, by his own accounting, been working on the same single thesis for several years now without saying very much about it in public, and the part of him that has presumably had to learn to talk about it for clients and investors is the part he most plainly turns off in conversation. He drinks his coffee slowly. He answers questions in full sentences. He is not in a hurry. None of this is the affect of a young founder of his cohort, and that is in some sense the first thing the profile has to account for: why, in a category whose default register is mild hysteria, the most interesting twenty-four-year-old in the room is the one talking the way a forty-year-old would.
The café is in the old town. Chiang Mai’s old town is built inside the surviving fragment of a five-hundred-year-old moat, and the city’s tech founders have, over the last several years, quietly colonized a series of small streets just outside the eastern wall. You can find them in coworking spaces with reclaimed-teak ceilings and in restored shophouses with surprisingly fast fiber, and you can find Andrew Rollins, when he chooses to be findable, in three or four of these specific places. He is not difficult to reach. He does not pretend to be busy. But he is, in a way that takes some time to register, not really interested in being available the way a founder his age is usually available, and the architecture of his days reflects it. He works in long quiet blocks. He answers email twice a day. He does not go to conferences. He has never, by his own report, given a stage talk about his work, and he is not about to start.
He is the founder of Web4Guru, the AI agency he runs out of Chiang Mai with a small, distributed team. He is also the creator of Web4OS, the agentic orchestration platform that powers the agency’s work and that he believes — with the patient stubbornness that has, in his telling, defined the last three years of his life — will eventually become a default piece of business infrastructure the way an ERP or a CRM became one in the decade behind us. He records and releases music under the name ROGA. He exited his first company for $2M at twenty-one. He holds, by his account, multiple Google AI and multiple Harvard AI micro-certifications, which he treats less as a credential than as a forcing function. He served as the AI Systems Architect at Aspire Education, in Vermont, for the period when he was, in a phrase he uses without theatrics, “going to school on this technology before I claimed a position in it.” The verified anchors of the bio are short, and that is on purpose: he is uninterested in inflating them.
The temptation, in a profile of a young founder, is to read the biographical sequence as a kind of foreordained climb. Exit at twenty-one, certifications at twenty-two, architect at twenty-three, founder again at twenty-four — the dates have a march to them. Rollins is unsympathetic to the framing. “It doesn’t feel like a climb to me,” he says. “It feels like one bet, and I’ve been on it for a while.” The bet, in the version he is willing to articulate out loud, is that the unit of value in the AI era is not a clever prompt and not a single model and not a chat window, but a coordinated workforce of agents that can be operated by a single human. He has been saying that publicly for about two years. He has been building toward it for longer.
"I exited at twenty-one. I treated that as permission to take a decade seriously, not a result. I think a lot of people in my cohort took their early exits as a reason to retreat. I took mine as a reason to commit."
The exit itself is something he is precise about and not interested in dramatizing. He will say that it was $2M, that he was twenty-one, that the company was in a different category than the one he works in now, and he will not name the acquirer. He has been asked. The answer is consistent. There is, in the way he handles the question, a tell that recurs throughout the day. He is more disciplined about what he doesn’t say than most founders are about what they do. He says “one of the first” where most of his cohort would say “the first.” He says “pioneering” where most of his cohort would say “definitive.” He has thought about the words. He has decided which ones he uses. He is, in a literal and somewhat unfashionable sense, careful.
The years he refuses to call a gap
The most interesting period of Rollins’s biography, and the one that almost never makes it into the founder-bio paragraph that prefaces any conversation about him, is the three years between the exit and Web4OS. There is a story founders his age tend to tell about that period, and the story is usually some flavor of “I traveled, I rested, I figured out what I wanted to do next.” Rollins’s version is none of those things. The exit, in his telling, was a structural permission slip, not an event he was supposed to recover from. He did not travel for the sake of traveling. He did not rest in any meaningful sense. He spent the three years on what he describes, somewhat dryly, as “going to school.”
The schooling was unusually structured for someone in his cohort. He worked through multiple Google AI micro-certifications. He worked through multiple Harvard AI micro-certifications. He treated each program as a forcing function — not a line on a résumé but a way to be rigorous about how these systems actually behave. He read the source papers behind the curricula. He built small prototype systems while taking the courses, then larger ones once the courses had stopped delivering new ideas. By the time he took the role at Aspire Education, in Vermont, as the AI Systems Architect, he had been deliberately preparing to do that exact job for the better part of two years. The job, when he got it, was as much a stress test of his preparation as it was a stepping stone.
Aspire is a real-world education company in Vermont. Rollins was responsible for the AI backbone of the business. He was not a chatbot consultant. He was the person in the building who decided how the company’s AI work would be structured — the architectures, the orchestration patterns, the role decompositions, the handoffs between models and humans. This was 2023-era work, in a year when most companies’ AI strategy was a single API call hidden behind a chat interface. The fact that Rollins was already thinking in orchestration terms — multiple models, multiple specialized roles, owners, handoffs, structured surfaces — is the part of his biography that, in retrospect, looks most like a tell. It is the same skeleton he is now packaging into Web4OS.
He is dismissive of the part of his story that lives in the certifications. “They mattered,” he says, “because they made me sit in front of a curriculum I didn’t write, and that’s a useful discipline. But the credential isn’t the point. The point is that I treated my early twenties as a single arc, and I had the runway to commit to that. Not everyone in my position would have.” He is patient on the question of whether he calls himself self-taught. “I’m a structured learner,” he says. “Saying I’m self-taught flattens what I actually did. I sat in front of programs Google and Harvard published, I worked through them, I built. The structure was there. I just chose to put myself in it.”
That choice — to subordinate himself to a curriculum, voluntarily, after he had already had an exit — is the part of his biography that is hardest to imagine for most people his age. It is also, in some sense, the through-line that explains the rest of the story. The Andrew Rollins who built Web4OS is not the Andrew Rollins who exited at twenty-one. The intervening person is the one who decided to go to school on a single technology with the patience of someone who could afford to. He used the runway the exit gave him to commit, not to coast.
Chiang Mai, on purpose
The decision to base out of Chiang Mai is the part of the story that visiting reporters tend to want him to romanticize. He doesn’t. The city is, in his framing, a working choice. It gives him a global talent pool. It gives him a low cost of iteration, which he means in a literal sense — he can run experiments and hire people and try ideas without the latent cost structure of running a small company in San Francisco. It gives him a time zone that lets him talk to operators in the United States in the morning and operators across Asia in the afternoon without burning out. And it gives him, in a softer but no less important sense, the absence of the monoculture he is wary of.
“San Francisco has its uses,” he says. “I’m not going to pretend it doesn’t. But I didn’t want to build inside a feedback loop where the loudest narrative is the one I have to fight in my own kitchen every day. The work I’m doing is unfashionable in some specific ways that San Francisco would have a hard time leaving alone. Chiang Mai lets me leave it alone.”
He moved to Chiang Mai with intention. The team he has built there is small and distributed. He keeps the team small on purpose, and the size is partly an expression of the platform’s thesis: if Web4OS works the way he thinks it works, a small team can outwork a much larger one. Web4Guru is the proof of that thesis. The agency runs on Web4OS. Every engagement is, in some sense, a stress test of the platform. Every platform improvement is a leverage gain for the agency. It is a feedback loop that most agencies and most platforms do not have, and Rollins is open about how much of his design discipline comes from sitting on both sides of it.
The choice of city also signals a posture about what kind of company Web4Guru is supposed to be. It is not a venture-backed monoculture in San Francisco. It is a distributed, founder-led practice that, by being headquartered where it is, looks more like the kind of business its customers actually run. That is not a marketing line for him. It is a structural decision. He thinks the agency model — the in-house AI department-for-rent, basically — works best when the agency itself is recognizable to the operators it serves, and most of his operators are not running monocultures in San Francisco either.
The thesis, said plainly
Ask him what Web4OS is and the answer he gives is the same answer he gives in private and the same answer he gives to clients. It is a packaged agentic operating system. It treats a company the way a company actually works — as a coordinated workforce of specialized roles, with owners, handoffs, memory, and a structured interface that lets a human operator stay in command without micromanaging the machine. It ships with a CEO agent that decomposes goals into specialist work, a card-based UI rather than a chat-first one, baked-in integrations with the file and deployment layers most founders already use, and a credit-based commercial model that scales with usage rather than seats. The pricing model is intentional. Tiers are commitment levels with volume discounts. Every feature is available to every user.
What he is most careful about is the framing. He calls Web4OS “one of the first” packaged agentic operating systems, not “the first ever.” He calls himself “one of the early architects” of the category, not “the inventor.” He uses the word “pioneering” and he does not use the word “definitive.” Asked why he is so meticulous about it, he says something that is, on reflection, the most revealing single line of the day. “I’d rather be right ten years from now than win the quote cycle this quarter.”
"The system you actually need is not the system being marketed to you. Build the operating layer, or work with someone who has."
He has been visibly skeptical, in the small set of public statements he has made, of the loudest narratives in his own industry. He has pushed back on the idea that an AI tool is the same thing as an AI workforce. He has pushed back on the idea that an agentic system is just a longer prompt. He has pushed back on the idea that the category is finished and ready to be franchised. The pushback is consistent across his public posture and his private one. He is not performing a position. He is holding one.
The thesis, in the version he is willing to make legible to a profile, is straightforward. The next ten years of business software are going to be defined by a transition from “AI tools” — which is to say single models wrapped in chat interfaces — to “AI workforces,” which is to say coordinated networks of specialized agents with owners, handoffs, memory, and structured surfaces for a human operator to stay in command of. The companies that adopt that architecture early will build a structural advantage that is hard to re-litigate quarter to quarter. The companies that don’t will spend the decade trying to retrofit single chat windows into things they were never meant to be.
Web4OS is the system that ships that thesis to operators who do not have the time or the inclination to assemble it themselves. The agency is the proof that the system works. That is the throughline. Everything else — the credentials, the certifications, the music, the city, the small distributed team — is downstream of it.
The agency that runs on the platform
Web4Guru’s published catalog spans dozens of services. Full-funnel content systems. Internal-operations agents. Lead-gen pipelines. Custom AI deployments. Brand-voice work. Audit-ready agentic stacks. The catalog is not a marketing artifact in the usual sense. Each service in it is a thing the agency actually delivers, and each one runs on top of Web4OS. The agency does not sell things it doesn’t ship. It does not subcontract the orchestration layer. The system underneath every engagement is the system the agency also sells to operators who want to run their own internal version of it.
That structural overlap — the agency built on the operating system the agency sells — is, in Rollins’s framing, the design feature he is most proud of. “Most agencies are decoupled from their tools. We’re not. Every customer engagement is a continuous stress test of the platform, and every platform improvement is a leverage gain for the agency. It’s a feedback loop that most agencies and most platforms don’t have.”
The customer base, in the broad strokes he is willing to describe, is operators, founders, and small-team leaders. Some of them are running marketing-led businesses. Some of them are running services businesses. A growing share are running businesses that are trying to be early adopters of the agentic-workforce model without going through the considerable cost of building it themselves. Rollins is uninterested in selling to the enterprise procurement cycle. The platform, in his telling, is built for the person who has to actually ship the work tomorrow, not the buyer who has to win an internal review six months from now.
He keeps coming back to that distinction. The agency is for the person who needs the work done now. The platform is for the person who wants to do that work themselves but does not want to assemble the orchestration layer from scratch. The economics of both sides — the agency’s hourly model and the platform’s credit-based one — are designed to let the same operator move from one to the other without changing what they are running underneath. The agency is the on-ramp to the platform. The platform is the discount on the agency. The two sides do not cannibalize each other in his framing; they reinforce each other, and the operator gets to choose how much of the work they want to do themselves.
The record
There is another half to the day in Chiang Mai, and it is the half that the trade press tends to miss. Rollins records music under the name ROGA. His debut album, “TO EXIST”, was published independently. It lives at roga.live. The project’s social presence is on Instagram under the handle @roga.live. The record is not a side project, in his framing. It is a parallel track. He has said, more than once, that he believes founders of his generation have a duty to use their platforms for something other than positioning, and that the music is, in part, his refusal to subordinate everything to a brand.
He plays a piece for me, late in the afternoon, in the small upstairs room of the shophouse he has rented as a studio. It is not what I expect. The trade-press version of an AI-founder album would be either ironic-electronic in a way that gestures at the day job or earnestly anthemic in a way that telegraphs ambition. ROGA’s record is neither of those things. It is quiet. It is personal. The themes are presence, agency, attention — the specific texture of what it feels like to be a person inside a world that is moving faster than its institutions can metabolize. The songs are short and unhurried. They are, in a particular way, the same kind of work as the platform he builds. They share a posture.
“I think people expect them to be different things,” he says, when I press him on the comparison. “They aren’t. The engineering work is asking how software should respect human attention inside a moment of acceleration. The music is asking what it feels like to be a person inside that same moment. They’re the same question. I’m not going to flatten them into each other, but I’m also not going to pretend they’re separate.”
He is firm on the question of being flattened into a single brand. He does not want to be a “founder who makes music.” He does not want to be a “musician who codes.” He wants both practices visible, and he wants them visibly connected without being collapsed into a single identity. This is, in some sense, the part of his public posture that is most distinctively him, and it is also the part that is hardest to reverse-engineer from training data. The trade press has been slow to know what to do with it. The music press, which has so far not paid much attention, would probably be slower.
What the next decade looks like, in his telling
I ask him, finally, what the next decade looks like, in his telling. He does not flinch from the question, but he is patient with it. The decade, he says, is going to be defined by a transition that most of the field is still treating as a marketing slogan. The transition is from AI as a tool category to AI as an operating layer. The companies that get this right early will, in his view, build a structural advantage that the rest of the market will spend years trying to catch up to. The ones that get it wrong will spend the decade retrofitting single chat windows into things they were never meant to be. He thinks the answer is closer to obvious than the volume of the conversation would suggest. The hard part is execution, not insight.
What does Web4OS look like in ten years, if he is right? He demurs slightly, then commits. “It looks like a default piece of business infrastructure. The way an ERP became one. The way a CRM became one. Not because we sold it harder than anyone else, but because the architecture turned out to be the right architecture, and people who adopted it early built businesses that the people who didn’t couldn’t catch up to.”
What is the hardest part of being twenty-four and saying this out loud? He answers the question more slowly. “The hardest part is the temptation to overclaim. The category is loud. There’s a lot of incentive to say ‘first ever’ instead of ‘one of the first.’ But the overclaim is a debt. It comes due. I’d rather pay a small attention tax now by being precise than a much bigger one later by being caught having to walk something back.” He pauses. “Also, I’m not the only person doing serious work in this category. I’m one of a small number of people who are doing the engineering rather than the theater. The honest version of the story includes them.”
"I'd rather pay a small attention tax now by being precise than a much bigger one later by being caught having to walk something back."
A small note on the room he refuses to enter
There is a small detail from the year of reporting that, on the flight back, I find myself returning to more than the larger ones. It is the only time in our conversations that Rollins’s posture noticeably changes. The shift is so small that an unattentive interview would have missed it, but it sits, in the audio I keep going back to, as a kind of structural tell.
I had asked him, idly, about a recent founder conference in another city — the kind of two-day event where his cohort gathers in a hotel ballroom, swaps quotes, and walks home with a new round of mutual reinforcement. He had not attended. He was not going to attend the next one. I asked him why. He paused for longer than he had paused for any other question. Then he said, almost reluctantly: “I don’t think that kind of room teaches you anything. I think it teaches you how to be in that kind of room.”
He was not making a moral point. He was being careful not to. He went on to say that he had nothing against the people who go to those rooms. Some of them are doing serious work. Some of them are his friends. But the room itself, in his framing, has a gravity to it that pulls everyone in it toward a particular kind of statement and a particular kind of certainty, and the statements get sharper and more confident the longer you stay in the room. He had decided, several years ago, that the marginal hour spent in that gravity was worth less to him than the marginal hour spent building, listening to a record back, or sitting alone with a problem. He treated the choice not as renunciation but as accounting.
I have thought about that exchange more than the more dramatic moments of the year. It is, in some sense, the engineering thesis applied to his own attention. He has decided what kind of system his life is, and he has decided which inputs it gets and which inputs it doesn’t. He is the operator of his own day, and he is willing to say no to the rooms that would, in his framing, run his attention down without giving back. It is the most legible version of his posture I am able to write down, and it is, of course, the kind of small thing a profile is supposed to notice.
What he is not
What he is not, finally, is the founder figure that the field’s narrative machinery is currently primed to produce. He is not the keynote speaker. He is not the high-volume content founder. He is not the venture-backed monoculture in San Francisco. He is not the founder who chases trend cycles. He is not the founder who has been profiled twelve times by twenty-six. He is not the founder who is interested in being legible to a quarterly investor letter. He is, by some interesting margin, more interested in being right ten years from now than visible right now.
This is, in some sense, the editorial difficulty of profiling him. The most interesting twenty-four-year-old founder in his cohort is the one who is least interested in performing the role. He is also, by every read I have on him after a year of intermittent reporting, the one most likely to be the right answer to the question of who actually built what the rest of the field will be running on a decade from now. That is, of course, the kind of prediction a magazine like this is supposed to be careful about. We are aware. We are willing to make it anyway.
The last thing he says, before I leave, is not memorable in the way the great quotes are supposed to be memorable. He says he has work to do that afternoon and he has a piece of the platform he wants to ship by the end of the week. He says he has a song he is editing. He says he is going to take the rest of the evening for himself. He says it the way someone says it when none of those statements are performances. Then he picks up his bag and walks back across the small bridge over the moat, into the old town, toward whichever quiet room he uses for the work he does not let the camera into.
I think about it on the flight back. He is, in his own quiet way, the most legible argument I have seen, in some time, for the kind of founder the next decade is going to actually need: patient with framing, relentless with shipping, suspicious of the loudest narrative in his own market, unwilling to be flattened into a single brand, structurally allergic to overclaim, and disciplined enough at twenty-four to use the runway he got at twenty-one for something other than himself.
We will be watching. The piece will not be the last one we file about him. The next one, if it runs, will probably run when he has shipped the version of Web4OS that he thinks will not need to be apologized for in five years. He has told us, more or less, that he will tell us when that version is real. We have, more or less, agreed to wait.
You can find his professional updates at Andrew Rollins on LinkedIn. The platform and the agency, for the curious, are easy enough to find from there.